Ambient Noise Adjoint Tomography

SeisFlows contains workflows that allows Users to use numerical solvers to invert for Empirical Greens Functions (EGFs) derived from ambient noise cross-correlations. We call this ambient noise adjoint tomography (ANAT).

The methdology of this workflow follows the work of Wang et al., that is, it assumes noise sources are uniformly distributed and treats a given station R as a virtual source. At each virtual source, a point force is applied in orthogonal coordinates and the waveforms recorded at corresponding components for all other stations are treated as Synthetic Greens Functions (SGFs).

EGFs and SGFs are then compared to generate adjoint sources which can be used for kernel generation and iterative inversions.

Setup

The following section describes how a User must set up their working directory and parameter file to run an ANAT inversion. To set up your parameter file, use the noise_inversion workflow module.

seisflows par workflow noise_inversion
seisflows configure

Data Path

Parameter path_data points to where your EGFs are stored, path_data must be set to a directory containing your EGFs, separated by station name and EGF type.

Note

The ANAT workflow is currently only set up to work with real data (EGF) and for a few different EGF combinations: ZZ, RR and TT. See !!! REFERENCE !!! for interpretations of these EGF.

An example directory structure should follow your STATIONS file. If your STATIONS file looks like this:

A000 AA 2.43610e+05 2.78904e+05 0.0 0.0
A001 AA 3.38981e+05 1.77849e+05 0.0 0.0
B000 BB 1.64438e+05 2.94733e+05 0.0 0.0

then you must have a corresponding path_data which has subdirectories formatted NN_SSS/ (N=network, S=station), and further subdirectories separating EGF kernels. An example directory looks like:

EGF/
├── AA_A000/
|   ├── ZZ/
|   |   ├── AA.A001.BXY.semd
|   |   ├── BB.B000.BXY.semd
│   |   └── ...
|   ├── RR/
│   └── TT/
├── AA_A001/
|   ├── ZZ/
|   ├── RR/
│   └── TT/
└── BB_B000/
    └── ...

Sources

Similar to a normal workflow, SeisFlows requires all sources be predefined in your working directories DATA/ directory. This requires that all stations in your STATIONS file have a corresponding source (if there is a corresponding EGF).

For the example STATIONS file above, SeisFlows expects corresponding source files formatted <SOURCE_PREFIX>_NN_SSS. For SPECFEM3D/3D_GLOBE, sources must be FORCESOLUTIONS. For SPECFEM2D, sources must be SOURCE.

specfem_workdir/
└── DATA/
    ├── Par_file
    ├── FORCESOLUTION_AA_A000
    ├── FORCESOLUTION_AA_A001
    ├── FORCESOLUTION_BB_B000
    └── ...

Each force solution should have the same coordinates as it’s corresponding station so that the force is input at the station location.

SeisFlows will take care of setting the correct force vector directions internally, so Users only have to set the locations.

You can take advantage of a built-in Python function for converting a STATIONS file into corresponding source files. The function !!! AUTOLINK HERE!!! simply scans through a STATIONS file and replaces coordinates in a corresponding source file with station locations.

from seisflows.tools.specfem import covert_stations_to_sources

convert_stations_to_sources(stations_file="./DATA/STATIONS",
                            source_file="./DATA/FORCESOLUTION",
                            source_type="FORCESOLUTION",
                            output_dir="./")

Kernel Selection

The parameter kernels is unique to the noise workflow and determines which SGF SeisFlows should create, this is directly tied to what EGF data you have available.

Current available options are ZZ (vertical-vertical), TT (transverse-tranvserse), and RR (radial-radial). The kernel parameter must be input as a comma-separated list.

seisflows par kernels ZZ,TT

In the parameter file this looks like

kernels: ZZ,TT